International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
LASER DATA FOR VIRTUAL LANDSCAPE GENERATION
Norbert Haala, Claus Brenner
Institute of Photogrammetry
University of Stuttgart
Germany
Norbert. Haala@ifp.uni-stuttgart.de
Commission III, Working Group 5 and 2
KEY WORDS: building reconstruction, laser scanning, urban models
ABSTRACT
Within this article a procedure for the automatic generation of 3D city models is presented. For that purpose existing ground plans
provided from a 2D GIS and height data measured by airborne laser scanning are combined. The presented approach can be applied
for a fully automatic reconstruction of an urban model, for interactive control and correction of these automatically generated results
and for a semi-automatic data capture. The results are 3D CAD models of the buildings. By an optional mapping of terrestrial or
aerial images onto the reconstructed facades, virtual city models can be generated.
1 INTRODUCTION
One of the main tasks in photogrammetry is the efficient data
capture for the generation of realistic representations of the real
world. A great impact on the future demands to these
geographic data bases can be expected due to the great success
of the World Wide Web. Amongst other things, this success
also resulted in a wide spread of international standards for the
description of 3D worlds like the Virtual Reality Modeling
Language (VRML). Additionally supported by the free
availability of VRML-browsers, 3D visualization is now
feasible for each standard user. Thus, the spread of applications,
which are based on the access and visualization of virtual 3D
worlds will increase tremendously. Possible scenarios are
tourism information systems, which can help the potential
visitor to plan trips as a preview afore his actual journey using
virtual reality techniques. Hotel reservation systems can be
integrated in order to give the user an impression of the hotel's
surrounding in a virtual 3D environment. Due to the increasing
power of personal digital assistants (PDA) also systems for a
personal navigation are imaginable, where the user's route is not
only overlaid to a 2D map, but is presented much more realistic
in a 3D virtual model.
All these applications will result in a great demand for the
production of virtual city models and landscapes, presuming the
availability of tools for efficient data capture. The development
of these tools has been in the focus of a number of research
activities and has led to a number of automatic and semi-
automatic approaches (Gruen, Baltsavias and Henricson, 1997).
Even though fully automatic procedures have been applied
successfully for a number of individual data sets, it is generally
agreed that their employment in a production environment can
not be expected in the predictable future. Consequently,
research and development has been concentrated to semi-
automatic procedures. A second common trend is the
integration of multiple data sources in order to make use of
different and complementary types of information. In our
approach an automatic data collection for virtual landscape
generation is enabled by the combination of Digital Surface
Models (DSM), which for our purpose is obtained by airborne
laser scanning with a 2D GIS.
Despite the high quality of laser DSM the automatic
interpretation of this data in order to reconstruct objects like
buildings, which are represented in that data set, remains a very
pretentious problem. In our approach this interpretation is
supported by the additional use of already existing ground
plans. Since these groundplans implicitly contain information
acquired by a human operator, the procedure can be
denominated as semi-automatic approach. On the other hand, if
the algorithm has access to the ground plans and the laser DSM,
the 3D city models are generated without any further
interaction. In that sense the procedure can be denominated as a
fully automatic approach.
2 ACQUISITION OF VIRTUAL CITY MODELS
Within the approach the reconstructed buildings are represented
by general polyhedrons. This presumes that all buildings can be
described by straight lines and plane surfaces. In order to
capture these polyhedral building models two major steps have
to be performed.
e A 3D measurement, which — in our case - is based on
height data provided by airborne laser scanning.
Alternatively, 3D points or lines can be derived from stereo
imagery.
e A structuring, i.e. the acquisition of topological relations
between the measured 3D primitives in order to get a
CAD-like representation of the reconstructed objects. This
structuring also implies a semantic attributation in order to
separate the captured primitives into different object
categories. In addition to the separation, a selection of
primitives, which are finally used for the description of the
respective objects can be necessary.
Frequently both steps are combined in a common process. In the
algorithm described by Gülch, Müller and Läbe (1999) basic
primitives are used for the representation of the buildings,
which are selected and approximately fitted to the image by a
human operator. These building primitives, which are also
applied in our approach, already define the topological structure
of the measured 3D points and lines. In contrast to that in the
approach described in this paper both steps are separated and
based on different data sets. For the required 3D measurement
